If you use nvidia-container-runtime, this is worth your attention — especially if JuiceFS is also in the picture. 1. An Alert Out of Nowhere, and I Panicked Over the weekend I was studying TensorRT LLM, and while I was at it I installed Dragonfly on the largest production cluster. Then

1. Introduction to Dragonfly The Dragonfly documentation at https://d7y.io/zh/docs/ already covers things in detail. Here is just a brief introduction to the main components of V2: Manager, which provides the UI, user management, cluster monitoring, task management, and other features Scheduler, which schedules traffic between Peers and provides preheating and other features Seed Peer, the back-to-source node used to download data from origin sites (Harbor, Docker.

1. What Is EnvoyFilter EnvoyFilter is a CRD resource in Istio that allows users to modify Envoy’s configuration to satisfy their customization needs for different scenarios. 1 2 3 4 5 6 7 8 9 kubectl get envoyfilter -A NAMESPACE NAME AGE istio-system add-request-id-into-ingressgateway 54d istio-system compression-gzip 18d istio-system custom-access-log 3d istio-system ingressgateway-settings 52d istio-system preserve-request-header-us-test-ingress-gateway 95d istio-system preserve-x-request-id 54d When using Istio, you will usually end up using at least a few EnvoyFilters.

1. Symptoms The business side reported that the API of application app-a was slow. Looking at the logs, one particular Pod was slow, and deleting that Pod so it moved to another node fixed it. From the monitoring metrics you can see that the Pod’s CPU usage did indeed spike sharply.

1. Subnet Management by kube-controller-manager kube-controller-manager has many controllers; the one related to Pod IP is NodeIpamController. The NodeIpamController mainly manages the podcidr of nodes: when a new node joins the cluster, it allocates a subnet to that node; when a node is deleted, it reclaims the subnet. The subnets